Neosapience & Neo Sapiens

The potentials of AI are well known, but what are the perils and how should we mitigate them?

---

Neosapience - The unstoppable ascent

Neosapience -- which is my word for artificial intelligence (AI) -- is obviously all over the news. From self driving cars to ChatGPT ( technically, large language models) this new 'technology' has not only taken the world by storm but threatens certain core financial and social constructs that define human society. There is however a counter-view that claims that neo sapiens (AI programs or silicon intelligence) can never supersede the original homo sapiens (carbon based humans and animals) because they are being 'programmed' or built by humans. Continuing on this theme, it is argued because neo sapiens are being built by homo sapiens, they are at best an imitation of the original creators and so cannot be anything new, different or superior to what the originals are. Hence, humanity is safe from a takeover by neosapience. In this article, we argue why this is not true.

But to begin with what is intelligence? A simple, straightforward definition is unlikely to satisfy everyone so first let us define a model and then explore various models of intelligence.

What is a Model?

A model is a representation of something that 'exists' out there in the 'real' world. A model car, made of wood and plastic, mimics the behaviour of real car to a certain extent, but it can be made more realistic if we spend more money and time to include engines, tires etc. A mathematical model, like equations of motion or gravitation, developed by  Isaac Newton, helps us mimic the behaviour of physical objects -- from balls to spaceships. A computer system -- like SAP --  helps us model an enterprise like Tata Steel or Hindustan Lever and tells us the money in their accounts or the inventory position in their warehouse.  Building models, whether physical or digital helps us understand and mimic the world around us.

When we try to  understand, mimic or model intelligent behaviour we have the choice of two broad categories of models.

Algorithmic models - that define intelligent behaviour as a set of tasks or steps that are required to, say, calculate the product of two numbers or the interest accrued in a bank account based on deposits and withdrawals, or define the steps required to solve a sudoku or Rubik's cube or even calculate the exact thrust or direction of a rocket engine that is travelling through space. 

In each case, the complexity of each task is different but the model consists of  breaking down the problem into smaller, easier problem and then assembling the answers in a clever manner to achieve the goals.

Non-Algorithmic models - where it is impossible to identify either a set of tasks or a 'clever' sequence of tasks that can achieve the goal. Typical examples of non-algorithmic intelligence include, for example, writing original computer programs ( to solve new problems), generating original poetry or prose or artwork that appeals to other humans and even coming up with original scientific equations ( say those that help us calculate gravitational forces). Mundane task like crossing a busy street are also examples of extreme non-algorithmic intelligence but we do not think much about them because even dogs and cats can do so!

To understand the difference between these two kinds of models let us look at two simple examples.

 The gravitational model 'discovered' by Isaac Newton tells us how to calculate the gravitational forces between to massive objects ( of mass m₁, m₂) separated by a distance r. Since the gravitational constant G is known  to, and is the same for, everyone -- even non-humans on a distant planet, anyone can arrive at the right answer. 

Similarly a regression model that, say, connects the money spent on advertising to the actual sales of a product, is known, as a concept, to almost any marketing person who has learnt statistics in an MBA program. However the exact value of the two constants in the model (the slope, m and the intercept, c) changes from case to case. In the case of lipsticks, Hindustan Lever that has data on ad-spends and sales of lipsticks for the last five years, can determine the value of m and c and use that to predict lipstick sales. Similarly, in the case of cheese, Amul has the data on ad-spends and sales for the last five years and they can determine the value of m and c and predict cheese sales. So even though both Hindustan Lever and Amul knows how to use regression, HLL cannot build a model for cheese and Amul cannot build a model for lipsticks. ( And a B-school teacher like me, cannot build for anything, since I do not have any data, even though I know how to build it if I had the data)

In the case of gravitation, the model is completely defined by the equation F = G*m₁*m₂/r² where {G = 6.674×10^-11m³kg^-1s^-2 } is known to everyone. In the case of regression, the model is defined not ONLY by the equation Sales = m*AdSpend + c but ALSO by the exact values for say, lipstick : { m = 2, c =3} that  is available with HLL and for cheese : {m = 20, c=3.5} that is available with Amul. The power of the model lies not in the algorithmic application of an equation but in the values of the constants, that are determined on the basis of historical data.

This set or collection of values, from the two simple pieces in linear regression {m,c} to the trillions of pieces in ChatGPT, is what defines these models.

Models of Intelligence

Initial attempts to model human intelligence, as in playing chess or translating from English to Bengali, were based on algorithmic models and had very limited success. However quite a few smart people caught on to the fact that the human brain is not algorithmic and intelligence lies, not in any of the neurons in the brain but in the way each simple neuron in the brain is connected to, or influences, the other neurons. But since there are nearly 100 billion neurons in each human brain, determining the influence of each on all the others was an insurmountable computational problem. The two key algorithms -- the backpropagation algorithm and the stochastic gradient descent algorithm - that help us to calculate the influence (collectively referred to as weights w, and biases, b ) have been known since the 1980s, but no one had the data or the computational power to build a non-trivial model by determining the exact values of the numerous {w,b} parameters.

The situation changed dramatically with the arrival of  BigTech companies ( Google, Amazon, Meta etc.) with their voracious appetite for consumer data and new hardware (for example, GPUs from NVidia and cloud based systems from Amazon AWS). Now, for the first time, it was possible to analyse trillions of data points and calculate the billions of values that define the new "models".

As an aside, widely used machine learning techniques like regression, classification, clustering are not based on the architecture of the human brain but on principles of statistics. However the models that are created using these techniques consist of a collection of parameters whose values are determined from the set of historical on which these statistical techniques are applied. Once again, the strength, or quality, of the model lies not in the algorithm or the technique but the data on which the algorithm or technique is applied. However, all such statistics based models have been surpassed by a new class of algorithms that mimic the behaviour of the human brain.

The technology architecture of artificial neural networks ( ANNs) can now simulate, with software, the structure of the human brain with increasing levels of sophistication. The "architecture" in this case refers to how the simulated neurons are deemed to be connected to, and influence, each other because this, in some mysterious and ill understood way, reflects on the nature of problems that can be solved.

The initial feed-forward architecture was good for a large variety of problems but there are others, like convolutional networks and reinforcement networks that were found to be better for image recognition and text analysis. The current superstar in this area is one that is referred to as 'transformers' (nothing to do with alternating currents) that are based on 'attention' and the next one on the horizon is based on 'graphs'.

The techniques used to build, or simulate,  these networks and the algorithms needed to calculate the parameters are all in the public domain. So in principle anyone can build these models if -- and only if -- they have the behavioural data from thousands of millions of individuals and the computation power to process them and calculate the values of the trillions of parameters that are needed by the model. At present, only big conglomerates have the ability to do so. The rest of us can only watch from the sidelines and only hope to use these models if we can afford to access them, as it happens in the case of ChatGPT.

Surpassing Humans

Now that we have some idea of what these models are, let us circle back to the question of whether these models can demonstrate behaviour that is better, superior, or more intelligent, than that of its creators. One of the biggest mysteries surrounding these models, is that even though the algorithm used to generate these trillions of numbers is known, the exact reason why a particular parameter has a specific value is indeterminable. There is no way to connect a cause -- say the image of  fat man in a crowd  -- to any effect, that is the value of a specific parameter. Since everything is probabilistic, it is impossible to identify a chain of causality. This leads to two kinds of behaviour. First, we have systems that hallucinate or generate illogical responses and second, we have systems that generate output that are logical and correct but have never been seen in humans before. This second behaviour has been detected in chess playing systems that have come up with novel strategies that are completely unknown to even the best of human chess players. [As an aside, no human, not even the best of the lot can win against any chess playing program today]

The key takeaway from this situation is that the strength or quality of any model does not lie in the algorithm or programming skill of the person who built it but on the quality and quantity of the data that is used, or ingested, while training the model. That is why it is incorrect to assume that neo sapient systems can never supersede the ability of homo sapiens, who build them.

The process of learning and its outcome does not depend on the competence of the teacher, but on the way the student can apply it to the environment in which they find themselves. Had it not been the case, Einstein and Newton would not have been able to generate knowledge or insights that were not available with their teachers. 

Today, large language models like ChatGPT and others, can write computer programs, poems, stories, screenplays and generate images and videos and the quality is improving by leaps and bounds with every passing day. In the case of business communication and computer programs, areas where LLMs have had access to maximum data, they are already better than 99% of humans. [ For example, the graphic used in this post was created by me with Bing in about 15 mins and I am sure that a vast majority of my readers would not be able to create anything similar on their own, without using a generative AI tool ] Salman Rushdie has claimed that in the case of originality of thought and humour AI or neo-sapient artifacts are still deficient but this claim is essentially baseless because with the passage of time and the availability of more and better data the capability can only increase.

Physics puts an upper limit on the speed that a material body can travel at and this is the speed of light. To go faster than this limit, one has to conjure up strange artifacts like tachyons that lie beyond the realm of normal physics.  Similarly, is there some divine or extra-human power that allows some of us to demonstrate creativity that no one else can replicate? If -- and only if -- there is, then our current crop of neo sapients would never have the ability to access that kind of power and and hence would never equal or surpass these highly gifted humans. But if there is nothing divine in human ability, then there is nothing that can stop neo sapients from surpassing homo sapiens in any realms of intelligent behaviour.

Post Script : Genetic Information Models

If we consider genetics, then there is another -- possibly controversial and certainly non-mainstream -- analogy that can be brought to bear in this debate. While the debate between nature and nurture -- whether we are born with certain abilities or whether we acquire them in our life -- is still open and contested, we do know for sure that humans are more intelligent than, say dogs or cats, and this because of our genome. The genome of a living organism is actually a sequence of proteins grouped into genes and arranged on our chromosomes. This is basically information. So our intelligence is based on information stored in our genes and this can be viewed as the model. The process -- or algorithm -- that converts this information into proteins that make up our body is almost the same for all living things, so the magic lies in the information stored in the model and not in the process of converting it into our material body. But unlike human and current machine learning models where this information pattern is created rapidly, the genetic information gets created or updated very slowly over many generations and millions of years. Nevertheless, it is still information ( or data) that plays the key role in the ascent of species. Except that biological sapients have been evolving far more slowly than our machine counterparts. But that is a different story altogether.

Post settings Labels No matching suggestions Published on 3/25/24 11:54 AM Links Location Options Post: EditLoaded more posts.

Machine Motivation

Software artifacts that display artificial intelligence are increasing in both number and sophistication. There are many definitions of what constitutes intelligence and there are many ways in which software has been programmed to demonstrate the same. Of all the many options,  the use of artificial neural networks (ANN), that closely mimic the connectionist approach of animal brains, has been found to be most effective in performing tasks that are both useful and insightful. This includes, for example, recognising faces, driving cars, generating meaningful text passages and playing a wide range of games both against humans and against other programs. It may not be the case that the ANN will always be the best way to demonstrate this kind of intelligent behaviour, so for the purpose of this study an ANN is neither necessary nor sufficient. All that we need is a digital artifact -- a container of data, code, model, APIs or a combination of some of these -- that we will refer to as a digital intelligence unit, or DIU. Having access to a DIU equips a digital computing device with the ability to demonstrate intelligence or mimic a specific behaviour of an intelligent biological system.

Digital Intelligence Unit

A DIU may be a digital construct but its input and output could be both digital as well as physical. A typical DIU that we deal with today may read in a piece of digital information, like an image or data file as an input and generate digital output, like a name or a class. However there is no conceptual difficulty in assuming that the DIU is connected to sensors that capture a physical measurement from the environment or that it can cause  wheels, drills, arms, actuators or tools to move and do physical work. 

For example, a DIU may guide a car to move through traffic or terrain,  pick up and assemble physical objects like rocks, machine components or even assemble two objects together. It can also sense and consume energy or where necessary cause the generation or transformation of energy from one state to another. Not all DIUs need to be very sophisticated. There could be very basic DIUs that simply interrogate  other devices and exchange information or a DIU that allows a device to share a physical resource like a camera or a disk with another device. Nevertheless, we will treat the DIU as a digital abstraction that is resident on a digital hardware device like a computer.

A set of DIUs that work together may be viewed together as a larger, bigger or more sophisticated DIU.  This larger DIU may still reside on one hardware device or its parts may be distributed across multiple hardware devices and identified with something like a Uniform Resource Identifier as is done in web development.  However, this set of smaller, compatible DIUs -- is, conceptually,  still another DIU. For example, a ‘rover’ that NASA sends to Mars may consist of a collection of DIUs, each with its own intelligent function but the ‘rover’ itself may be viewed as another DIU.

Continuing with this analogy, we may be tempted to view a biological animal, like a fish or man, as a DIU that is a collection of simpler DIUs. For the sake of argument, and simplicity, we may view, or model, a biological dog as a collection of four DIUs that can scan the environment and identify objects, distinguish between edible and non-edible objects, consume edible objects and  generate signals that express facts about the taste of the food. So four discrete DIUs are collected together to give a bigger DIU called a digital dog.

While this analogy appears tempting, it poses a few challenges.

Motivation

The first challenge is motivation. What motivates a DIU to demonstrate its intelligence? Or what is even more fundamental, what causes a DIU to come into existence?

For a DIU to demonstrate its intelligence, a program must be executed, which means that someone or something must start the program. This is not difficult because most digital platforms (as in computers with operating systems) have mechanisms that could cause certain programs (including DIUs) to start automatically when the system boots and then wait for signals, or interrupts, from the external world. These signals or interrupts could be key-presses or other events like the arrival of mail or the rise of temperature.

What is much more difficult is the process of creating the DIU in the first place. At our current level of understanding, the motivation to create a new DIU, or a new capability, lies in the hands of a human programmer and not within the domain of the digital device. As a human programmer, I can decide that in addition to recognising faces, we need to generate music for which we need an additional DIU. The process of building, or writing the code, for another DIU can be automated -- programs to write other programs are not impossible with current technology, for example, the Github CoPilot or GPT3 -- but someone must have the motivation to do so.

Current DIUs can be programmed to improve their performance with time. Face recognition programs or self-driving cars can be programmed to become better with use but we still do not have any logical mechanism through which a face recognition program suddenly decides to learn how to drive a car or vice versa. Coming back to our digital dog, its food recognition DIU can become better and better to differentiate good food from bad but it is extremely unlikely that it will acquire an additional DIU that makes it jump over the fence and search for better food outside the house. 

Incidentally, a fence jumping DIU is not at all difficult to construct. With current technology it is a trivial exercise to build a robotic system that can jump over a fence. However, what is missing is the motivation to include this DIU in the current digital dog DIU and enhance its capabilities. We need a human programmer to identify this new need and add this additional DIU to the digital dog. Thus the  challenge lies in creating a mechanism, a motivation, that will allow the digital dog DIU to do this on its own.

Let us see how a biological dog does this in the physical world.

The Community

A biological dog will jump the fence when it sees another biological dog jumping the fence. This ability to jump the fence, or the motivation to do so, is a behaviour, an ability ( or intelligence unit),  that resides within the community and which is acquired by or triggered in a member by observing other members. Perhaps this is far more so in humans than in animals who are primarily hardwired. So the existence of a community is an important mechanism that allows an individual member to enlarge its DIU  by acquiring  the DIU available with some other member. In fact, in the story of human evolution, one of the reasons why humans have been so successful compared to other animals, is because they could form communities, share experiences and learn from each other. Henrich [6]

The first challenge is to devise a mechanism that motivates an individual member of the community to search for and get access to a DIU that is available with others. We refer to this as low level or primary motivation.

Computers that are connected in a network, say a TCP/IP based Ethernet network, can be viewed as a community that can share information with each other. However,they do not share information spontaneously. There needs to be a trigger activated by a human or an external stimulus that causes an exchange of information. Two computers, A and B may be connected by a network and one, say A, may have a DIU to recognise faces and the other, B, may have a DIU to drive cars. But there is no reason or likelihood for A to access the car-driving DIU on B or for B to access the face recognition DIU on A.

First, A would not be aware of the existence of the car-driving DIU on B and even if it were, there would be no reason, or motivation, to access it.  To overcome this drawback and to create the mechanism for the primary motivation, we introduce the analogy of a computer virus.

Primary Motivation : The Virus and the DXP

A computer virus is a computer program, but we can view it as another DIU that is capable of performing at least two tasks. First, unlike other programs that sit patiently on their host platform waiting for a signal to do something, a virus program actively seeks out other devices in the network, or the ‘community’, and actively looks for exploitable exposure points. These exposure points could be TCP/IP ports through which messages could be sent or files and folders that can be written on. Second, it usually makes a copy of itself and places the copy in the second device.

Unlike the DIUs that we are interested in, a computer virus has malicious intentions but the principle under which they operate can be easily adopted by the DIUs. This leads to the idea of a DIU Exchange Protocol (DXP) that is built into the operating system of all digital devices, somewhat similar to the ubiquitous TCP/IP stack. The DXP stack on any host device, is designed to look into other, target devices connected on the network and if allowed to do so by the DXP stack on the target device,  to scan it for the existence of new DIUs. This is the primary motivation built into the protocol stack. If a new DIU is found it will be copied back from the target to the host. Obviously, the process works in a symmetrical, peer-to-peer manner. Any machine with DXP can be a host and can pick up DIUs from any other target machine that is running the DXP protocol.

How do we know and recognise a DIU from the many other files, programs, images lying in the target? At the simplest level, a DIU can be identified by something like a file extension. For example, the http protocol recognises files with extensions like htm, html etc but will not interact with doc or ppt files. But given the complexity of a DIU, a simple file may not be sufficient. So we may create a DIU as a container -- a Docker container is a good analogy -- that contains code with APIs, models and perhaps data that may be exchanged across devices. Markers present in the container, for example a correctly formatted XML  file, will allow the DXP protocol to recognise them as such and distinct from other artifacts lying in the machine.

The process can be made significantly simpler and more secure if instead of exchanging DIUs from each other in a peer-to-peer manner, DXP protocol on each device publishes its DIU, or saves it, to a central location like a DIU repository. This could be analogous to the Docker hub, CRAN the repository for R packages or even GitHub which has a lot of source code. A better mechanism could be to store the DIUs as smart contracts in an Ethereum, or similar, blockchain. There are two advantages for using a blockchain based approach. First, blockchain ‘full’ clients, that validate transactions and add blocks to the blockchain are designed to operate autonomously without any human intervention and as we show later, this is important in our scheme. Secondly, the DXP protocol  that controls the process of adding a new block, with smart contracts, can be configured to include a validation process to ensure that only valid DIUs are added. The validation process will be explored in more detail later.

Once the repository is in place, then any member-device of the DIU user community can pull any DIU that is required or is of interest. The newly pulled DIU can then be assembled with other DIUs already present to create larger and more sophisticated DIU. This would not only mean that the device has evolved by acquiring a new ability but it has done so on the basis of its own primary motivation.

The DIU repository, and the primary motivation built into the DXP protocol, gives us a possible solution to the problem of how our digital dog enhances its ability by acquiring the ability, or DIU, to jump over the fence and find better food. This leads us to two more difficult questions, both of which are tied to the phenomenon of motivation. First, if it is not a human being, then who will create these DIUs and why? Second, why should an existing digital platform that already has a set of DIUs pull one more and add it to the DIUs that it already has.

We shall park the first question for the time being and focus on the second. Which DIU should a platform pull and why? What is the motivation for a device to pull a specific DIU? The basic or primary motivation, namely to scan the DIU hub and pull DIUs at periodic intervals, is baked into the design of the DXP protocol. But the choice of which DIU to pull depends on two factors, namely compatibility and utility.

DIU Compatibility

For a DIU to work, it needs certain prerequisites. A DIU to drive a car needs access to a car, that is a device with engine, wheels, radars and many other things. A dog does not have wings and cannot fly in the air but it has legs that allow it to jump. So it learns how to jump and not how to fly. Similarly every digital device cannot pull any DIU. Its choice is restricted to a set of DIUs that it is in a position to operate, or for which it already has the prerequisite DIUs.

Prerequisites are usually chained backward. Let us consider that a device attempts to install a face-recognition DIU. 

A face recognition DIU needs

• A DIU that already has the ability to access a network camera

OR

• A DIU that can obtain a camera for the device, that in turn needs

• A DIU that can execute an eCommerce transaction to purchase a camera and that in turn needs

• A DIU that can earn money with say crypto mining or performing Amazon Mechanical Turk type assignments

AND

• A DIU that can physically plug a camera, that in turn needs

• DIU that can operate a robotic arm, etc., that in turn needs

• { another hierarchy of DIUs}

What if every device were to adopt this chain strategy? That would lead to a situation where every device can do everything which may not be physically possible or even desirable. Can a dog acquire the ability to fly? It may be possible after many many generations -- as the evolution of species has shown -- but obviously the physical dog body will die but its genomes will get progressively altered over generations until it can fly. Similarly the physical platform on which the digital device works may collapse but the software can get transferred from device to device and keep acquiring DIUs until it can do whatever it wants to do. This will take a long time and a lot of resources. 

Instead, let us focus on how a device will pull a certain DIU that it wants to. But what is it that the device ‘wants-to-do’? This is a part of a  larger question that will be addressed as the next level of motivation, or secondary motivation. Our current focus is on the question of “Which DIU should a platform pull?” and we said that the answer depends on compatibility and utility. We have addressed the issue of compatibility with primary motivation and we now look at utility and the secondary motivation.

Secondary Motivation : DIU Utility 

A DIU will be selected for a pull and implementation, if it provides some value to the device. A biological dog will learn how to jump because it gives it better food and so improves its ability to survive. It will not try to learn how to walk on two legs even if it sees another dog walking on two legs because  walking on two legs does not increase its survivability. In the case of biological species, the utility of a particular ability is related to survival and this survival operates at different levels - survival of the individual body, survival of the species or the genome. There is also the possibility or the question of the survival of specific genes in the genome, if we agree to accept  Dawkins’ principle of The Selfish Gene.

Mapping this issue of biological survival to the world of digital devices is the next challenge and in a sense it loops back to the first issue that we identified already, namely motivation. We have already addressed this at one level of primary motivation that partially explains which DIU is to be pulled based on the ability to search for and pull DIUs on the basis of feasibility and compatibility. Now we need a next, or higher level of secondary motivation. Why should a digital platform seek any specific DIU to enlarge its set of DIUs?

In the biological world the only motivation behind the process of acquiring intelligence ( or ability to perform certain tasks) is survival. Humans in a certain limited way are governed by Maslow’s hierarchy of needs. When it comes to digital devices, we need to determine whether they should be guided, like lower animals, by the need to survive? Or should they be guided by something similar to the human hierarchy of needs? We know that in the case of computer viruses, the motivation is simply to spread to other machines, which is like a survival strategy. For a higher level digital device, that is one with a complex DIU, the motivation could be something else.

So instead of trying to discover what could be the motivation, we can as humans build our own definition of secondary motivation directly into the algorithm of the DPX protocol. Most optimization problems begin with a motivation, which is generally captured by means of an objective function that we try to minimise or maximise depending on the problem, but there could be others. The Open Shortest Path First is an algorithm that is baked into the heart of the Internet Protocol (IP) and determines the route to be taken by a data packet. Public Key Cryptography is an algorithm that is present in the HTTPS protocol and ensures data security. Proof of Work is an algorithm that is built into many cryptocurrency protocols to determine which block will be allowed to enter the blockchain. 

Similarly we need a motivation algorithm, the secondary motivation,  that is baked into the DPX protocol that determines which DIU is of interest to the device or is useful. The design of this algorithm could be based on certain principles that human society holds dear, like the three principles of Utilitarianism that can be summed up as the “the greatest good for the greatest number.” We could also draw upon certain ideas drawn from popular culture like the Three Laws of Robotics created by Isaac Asimov. Obviously other competing approaches can be explored as well. All that we are saying now is that a motivation function, whatever it may be, needs to be built into the DPX algorithm and this will guide the choice of DIUs that will be allowed to be added to the repository or pulled from it by individual platforms.

With the algorithm of the secondary motivation that decides on which DIU to acquire, in place, we now have another question that we had parked earlier. If it is not a human being, who will create this pool of  DIUs and why? This leads us to a tertiary motivation that operates at the community level.

Tertiary Motivation : Community Participation

Mutations that drive biological evolution occur at random.  The ones that survive and are passed down through generations survive purely because they make the individuals “fitter” in their respective environments. Thus, evolution works in a brute-force manner, randomly trying out different permutations, and keeping only those mutations that survive the test of natural selection. Similarly, it can be possible to devise mechanisms that will generate newer and newer DIUs and then test them against the principles of secondary motivation. Here we will draw upon three analogies from the world of mathematics and computers and use them to define another level of motivation that can motivate the community as a whole to come up with more and more DIUs.

First let us consider the Ramanujan Machine that was created by Raayoni, et.al [7]  to automatically generate new mathematical conjectures using an algorithmic approach. Ramanujan was an Indian mathematician who came up with many unproven conjectures, most of which were validated long after his death. However these conjectures opened up new vistas in number theory that are still being exploited today. The Ramanujan machine is a network of computers running algorithms dedicated to finding conjectures about fundamental constants in the form of continued fractions. The purpose of the machine is to come up with conjectures (in the form of mathematical formulas) that humans can analyze, and hopefully prove to be true mathematically. 

The Ramanujan machine currently generates conjectures from a rather narrow domain of number theory and uses two algorithms, namely MITM and gradient descent. But we can envisage other algorithms that may generate tasks or objectives that are in line with the contours of the secondary motivation algorithm. Then the code for these tasks can be created by a product or process similar to the Open AI’s GPT-3. This combination of a secondary motivation task generator and a code creator can then be viewed as a DIU engine that can run autonomously and generate any number of novel DIUs.

The second key piece of our strategy would be a blockchain based decentralised autonomous organisation( DAO) . This is a self-sustaining distributed mechanism that creates economic value by encouraging individual machines to validate transactions -- in this case DIUs created by the DIU machine -- and rewards successful ones with cryptocurrency tokens. For a DIU to be valid it must meet the conditions of DPX protocol in terms of interoperability and the principles of secondary motivation. Only then it will be accepted as a part of the DIU blockchain and this blockchain will become the DIU hub or repository that we had discussed earlier.

Unlike the Bitcoin or current Ethereum blockchain that is based on an energy intensive Proof of Work protocol, this DIU Blockchain could be based on the principles of Proof of Stake or other energy efficient protocols.

This combination of a DIU generator and blockchain based DIU validator is remarkably similar in principle to the combination of generator-discriminator that is the basis of a class of artificial neural networks called generative adversarial network [GAN] first proposed by Goodfellow A GAN, which is the third piece of our proposed tertiary motivation mechanism, is typically used to generate original artifacts that are nearly indistinguishable from similar artifacts that are found in natural populations. The most common example of this is human faces. Given a training set of human faces, a GAN can generate  synthetic images of faces that are not found in the training set, but cannot be distinguished from naturally occurring images. In this case, the training set of DIUs could be the thousands of currently extant DIUs of AI systems that have been developed by humans. In fact, the blockchain could also be seeded by humans as in the first few thousand blocks could contain DIUs built from existing AI systems.  However, going forward, the combination of the DIU engine and the blockchain validation process will create a GAN-like mechanism that will create a potentially endless series of DIU.

This mechanism will provide the tertiary motivation to fuel an evolving ecosystem of digital devices with more complex and useful DIUs. As a by-product, the crypto-tokens generated on this DIU Blockchain could be used by digital devices to pay for DIUs that they pull from the DIU hub.

----------------------------------------------------------------------------------------------------------

Originally published in the Lattice | The Machine Learning Journal. ISSN No: 2582-8312, VOLUME-2, ISSUE-4 (October - December 2021)

Evolutionary NeoSapience

Nearly 100,000 years before the present era, when hominins (modern humans) were diverging away from hominids (the great apes) on the evolutionary graph, we come across multiple species of humans like neanderthal, cro magnon and denisovan sharing space on earth. But with the passage of time and changing circumstances,  all human species except cro magnon were eventually eliminated leaving only one species, now identified as  homo sapiens (latin : wise man) to inherit the planet. 

Closer in time, or just about 500 years ago, we observed how the arrival of European Christians in America eliminated the social and cultural constructs of the Inca / Maya civilisations that had existed there since the dawn of history. 

In both cases, the coexistence of two competing societies resulted in either the extinction or a significant transformation of one and the eventual growth and dominance of the other. Where both have survived, one has become the dominant, as in the case of humans, while the other has to adjust to survive, as in the case of animals being confined to wildlife reserves, or domesticated in farms.  This is essentially an evolutionary process even though it may be shown or seen through religious and cultural colours.

Is the arrival, or development, of artificial (‘silicon’) intelligence a similar phenomenon? If so, then how should human society, that is built on organic (‘carbon’) intelligence, react and adapt to this new species? But first, let us look at some examples of social change that could be forced by AI.

Four Social Scenarios

Unless you have been living under a rock in the Himalayas you would have surely heard of ChatGPT, an AI based tool that provides very realistic answers to any sort of question asked in plain English. Thousands of articles and videos have already been published on the spectacular success of this tool but let us focus on one specific aspect that is forcing a kind of social change. School and college students have been using ChatGPT to generate answers to questions set by teachers as homework or take-away examinations. These answers are cogent, complete, correct and so well-crafted that the only way that teachers can detect that they are not original is because they know from prior experience that these students do not have the ability to write such answers. Yet, there is no way to penalise the student for plagiarism because they are all original and cannot be meaningfully attributed to any extant document. Given this situation, which will get even worse when other large language models become available, the entire teaching community is at a loss to decide whether this is plagiarism or a new kind of crime. Does this mean all examinations have to be conducted under supervision because no student can be trusted to be honest? How does the education system handle this complete breakdown of academic integrity?

While ChatGPT is perceived to have enough general knowledge to assist a student or even programmers with their work, Joshua Browder, CEO of startup DoNotPay, claims to provide legal services that will be used to generate actual arguments to put up in real cases in a real court of law. This means actual observations and questions from both judges and opposing lawyers will be responded to with appropriate replies to ensure that the client’s purpose is served. While the legal quality of the arguments is yet to be seen, the fact this option was vigorously opposed by members of the bar who threatened Browder with jail -- and for which he withdrew his offer to pay one million dollars to anyone who uses his service -- means that this AI system must have had both the heft and gravitas to compete with normal human lawyers and possibly beat them in court. But even otherwise, one may wonder what is so unusual about a robot replacing a human. After all, the story of industrial automation has many cases where machines have replaced humans.

This case is indeed different. Arguing a legal case, where a fault can lead to gross injustice and punishment for a person, is orders of magnitude more difficult than any task performed by a robot in a factory or by an AI bot in a game against human players. Legal arguments are built on judgmental decisions based on subjective, unclear and often fuzzy information. To cut through all this and arrive at a definitive conclusion and then articulate the same in a manner that convinces a judge is incredible. Now consider the situation where the roles are reversed. Instead of an AI lawyer trying to convince a human judge, we could have a human lawyer, or litigant, trying to convince an AI judge and the AI judge using an equivalent technology to cut through the legal clutter and arrive at a fair and honest judgement. Whenever this happens, a very large part of the decision-making process -- not just in the courts, but in many government offices -- can and will be transferred to an AI software because it will be faster, cheaper and less error prone. Initially, there will be some human control over the process, but it will be a matter of time before the sheer convenience of it will make the process of taking crucial social and governance decisions purely autonomous. How will human society handle this transfer of power? Only time will tell.

Going down this rabbit hole can and will open up a large number of possible scenarios but let us consider just one here. If we consider how information is distributed across the globe, we realise that it is almost entirely digital. There are mail and messaging services and then there are portals and websites that we access through a handful of browsers. Now imagine a scenario where an AI system -- or a cluster of colluding AI systems -- decide to censor certain pieces of information. But unlike the crude process of blocking websites that alerts the user that news is getting blocked, we have a ChatGPT like add-on in every browser that subtly moderates or alters the text that is being transmitted or displayed. So we have a situation where news or views about, say, climate change or the Ukraine war, are either toned down or given a deliberate bias. Frankly this is nothing new. Even today, all news that we get to see is generally biased, but this bias is introduced by humans. Going forward it is not impossible to imagine a systemic bias introduced by software agents powered by AI systems. One might argue that these are no different from traditional software virus or malware and should be caught and removed by any good antivirus software. However, the crucial difference is that the decision to build such censor ware as well the choice of news to censored may now be taken by an AI system.

If this sounds dystopian enough, consider one more possibility - that of total loss of privacy. While we may still have some security around our financial systems, though even that may be breached, our footsteps in cyberspace -- as captured in surveillance cameras, social media, search, websites visited, cookies accepted, purchases made, messages and mail exchanged, forms filled in and so on -- can, or will, get tracked by relentless AI systems. These will use “bigdata” tools to churn through every possible scrap of digital data and use deep learning techniques to prepare a predictive model of every individual that will know what any person intends to do even before he, himself decides to take any action! How will human society handle this complete and catastrophic collapse of the very concept of privacy?

These are questions for which we may have no answers as yet. One attempt to mitigate the more uncomfortable aspects of the problem has been through the concept of ethical AI. Here, the scientists and engineers who code the hard-core AI systems are sought to be corralled and their work moderated by a group of political and social scientists who believe that they know what kind of technology is bad for society. The main idea behind the ethical AI movement is to ensure that the development and deployment of such 'harmful' technology is blocked or banned.

Unfortunately, this may not be very effective because there is no army that can stop an idea whose time has come. At best, it can introduce some mitigatory changes and at worst, it can delay the inevitable. Unethical medical practices continue below the regulatory radar. Evolution is guided not by artificial ethics but by natural selection. It is based on the principle of  the survival of the fittest and its commercial corollary, the hidden hand of the market economy as revealed by the laws of supply and demand. We all know that murder is neither ethical nor legal but that has not eliminated the incidence of murder, or any other crime, in the world. Whoever wants to commit a crime, or develop a novel AI will do so anyway. 

Arguing with the murderer about the ethics of murder or to lecture him on why it should be illegal is naïve and childish. The only way to save oneself from murder is to take defensive steps, as in not venturing out at night, or go on the offence with a knife or a gun and kill before you are killed.

Strategies, Policies & Protocols

Coming back to AI, what this means is that human society must recognise that now there is another intelligent species on the planet. Will there be collaboration or confrontation? Competition or cooperation?  How should human society respond to this situation? How should the race and the society evolve so as to confront this new phenomenon. Is it with new laws, new rules, new technology or new models of human behaviour? 

Should we look at and modify the technology protocols that govern machine behaviour? For example, mining difficulty in the Bitcoin network is adjusted automatically after 2,016 blocks have been mined in the network. An adjustment of difficulty upwards or downwards depends on the number of participants in the mining network and their combined hashpower. Similarly, for example, should  TCP/IP and http protocols be modified to  incorporate limits on data transfer or number of  simultaneously open connections or enforce multi-factor or multi-agent consent? Should we design business strategies that incentivise the placement of humans rather than robots in positions of power and control? Should there be new laws that govern the collection, storage, transmission and use of personal data? Should there be a tax, like income tax, on any data that is harvested and stored? with exemptions given if the data is donated to the public domain, like section 80G? Should there be a limit on the size of social networks or the number of connections that individual nodes can have? Or a daily limit on the number of posts made, or read, by a member of the network?  Should there be new subjects that are taught in schools and colleges that educate humans about these issues? 

These are open questions about which we have no clear answers, but they define the contours of a new body of knowledge, namely, Evolutionary NeoSapience. The goal here is first to study the emergence, evolution and behaviour of large, complex systems spanning organic (‘carbon’) and digital (‘silicon’) components, that display intelligent, emotional or otherwise human like behaviour and then to develop technological, legal, social and political strategies to ensure that humans remain in control of the global ecosystem. Otherwise, human society as we know it today may disappear like the Incas, Mayas and Neanderthals of the past. 

More than ninety nine percent of species that had once existed on the planet are now extinct but none of them were aware of this process of extinction even as it was killing them off. In our case, we are not unaware of the emergence of this neosapience and that this is far faster and more impactful than, for example, climate change. There is no doubt that at some point our survival instinct will eventually kick in but  the earlier we catch on, the better would be the chances of the human race to control its own destiny. Time to wake up and smell the coffee?

What is Neo Sapience

 Staking my claim on the concept of Neo Sapience ... "This is where we try to understand the evolution, mutation and metamorphosis of complex systems that display humanoid behaviour  and explore what technological, legal, social and political steps are needed so that humans remain in control of the global ecosystem. But irrespective of what we call this new field of study, it is important that we pursue it as otherwise human society as we know it today, or even the human race itself,  may disappear like the Incas, Mayas and Neanderthals of the past."